Pressure - Nutrient loading - Interlinkages

2.3 Interlinkages - using the DPCER approach

2.3.2 Pressure - Nutrient loading

Freshwater quality is dependent on wide range of physical, chemical, biological and societal factors, with the contribution of hydrology as a key element (Walter et al. 2000), since the hydrological distinctiveness of a catchment influences the dynamics of water flow. Understanding the various hydrologic processes involved in the transport of solutes and sediments to streams leads to a more successful implementation of water resource management guidelines and policies.

27 Hydrology

The hydrologic cycle is a complex system of interactions, and on a catchment scale involves atmospheric moisture (precipitation, evaporation, interception, and transpiration), surface water (overland flow, surface runoff, subsurface and groundwater outflow, runoff to streams and ocean), and subsurface water (infiltration, groundwater recharge, subsurface flow and groundwater flow). All these interactions direct the changes in the morphology and habitat of rivers and are intimately connected with climate, geology, topography and general catchment features (Molnar et al. 2002).

Evapotranspiration is the main process responsible for annual changes in water yield as a result of alterations in vegetation. Changes in vegetation caused by various land-use activities have been the focus of many hydrological studies, with four main categories of vegetation change – afforestation, deforestation, forest conversion and re-growth.

Depending on the activity, water yields from catchments are affected including changes on a temporal scale (Brown et al. 2005). Water yield changes can be the result of changes in surface runoff, in base flow or both. In West Africa, hydrological responses are non-linear and when thresholds are exceeded, lead to dramatic increases in water yield (Li et al. 2007).

Runoff response, where surface flow is generated when the soil is fully saturated, i.e., Hortonian flow (Horton 1933), or when precipitation exceeds the infiltration capacity of the soil, i.e., Hewlettian flow (Hewlett and Hibbert 1967), are strongly dependent on the initial soil moisture content, slope length, infiltration properties and saturated hydraulic conductivity (which determines the maximum capacity of soil to transit water). These are factors that also influence flow paths and rainfall runoff responses (Ajayi 2004). In the West Africa region, the generation of runoff after a storm is dominated by the Hewlettian flow, or ‘infiltration capacity excess overland flow’. This process begins when rainfall intensity exceeds the infiltration capacity of the soil such that runoff flows downhill towards the valley due to the combination of high intensity rainfall (>100 mm/hr, for between 10 and 30 minutes) and poor infiltration properties of the soils in West Africa (Ajayi 2004; van de Giesen et al. 2005). The rate of surface runoff is the difference between the rainfall intensity and the infiltration rate, assuming that evaporation during and immediately after the event can be considered statistically negligible (Shahin 2002). The

amount of runoff varies depending on the scale of investigation, with reducing discharge rates as scale and slope lengths increase. Field measurements have shown significant reduction in runoff of 40-75% on 12-m slopes as compared to runoff from 1.25-m slopes (van de Giesen et al. 2005). These differences are due to functions of rainfall duration and intensity, slope length and gradient, surface roughness and infiltration capacity – features important in the understanding of how modifications to land-use and cultivation techniques can influence water and solute transport at different scales. High nonlinearity of the rainfall-runoff response also contributes to the observed discharge, as in the Volta Basin, for example, 340 km³ of rainfall has to occur before a significant amount of runoff can be observed, with half of the rainfall generated as runoff when the threshold is surpassed (Andreini et al. 2000). Runoff into streams may occur only from some parts of the catchment, or source areas, which vary in performance during a storm event (Shahin 2002).

Modifications of any parameters involved in the hydrological processes can therefore have dire consequences on the entire cycle.

The hydrological regime of a catchment, essential to its ecological functioning, is significantly influenced by two main parameters – the nature of the flow (intensity and duration), and the stream flow variability (seasonal cycle, recurrence and predictability) (Molnar et al. 2002). Extensive ecological consequences can, therefore, result from changes to the regime caused mainly by anthropogenic activities such as deforestation, land-use changes, natural variability in large-scale atmospheric circulation patterns, and climate change due to increase in greenhouse gases and global warming. In Ghana, the hydrological effects of changes due to anthropogenic land-use, despite the global impacts of climate change, has been seen in the significant reductions in rainfall amounts in the humid zone of the south western rivers system (which historically has higher levels of land use change and deforestation) when compared to the semiarid northern savannah regions. With the higher annual variability of discharge (57%) than rainfall (7%) in the Volta Basin, for example, such changes in annual rainfall can lead to considerable changes in discharges (Gyau-Boakye and Tumbulto 2006). The conversion of natural land and forests into agricultural lands results in significant changes to hydrological patterns, such as higher water yields resulting from lower transpiration rates and reduced infiltration due to reduced forest cover,

with negative effects on soil fertility and water quality (Mungai et al. 2004). In addition, the practice of burning cleared vegetation in the traditional bush fallow system exposes the underlying soils, and changes physical and chemical composition of the soils. This increases sand proportions (Bagamsah 2005) and reduces the rate of infiltration of precipitation into the soil (Bijker et al. 2001) due to soil crusting (Mills and Fey 2004), further affecting surface runoff and stream hydrology.

There is a wide variety of hydrological models that estimate the hydrological processes within a catchment driven by interactions between climate, soil, vegetation and surface topography. These models are usually grouped under four main categories – (i) physics-based or fully distributed models, (ii) conceptual models, (iii) metric models, and (iv) hybrid metric-conceptual (HMC) models (Amisigo 2005). Physics-based models are the most complex, with partial differential equations representing all the component processes of subsurface and surface flow. These equations are based on the physics of the processes and the model requires the estimation of numerous parameters. Examples of the fully distributed models include the Système Hydrologique Européen (SHE) model (Abbot et al. 1986), the Institute of Hydrology Distributed Model (IHDM) (Bevern et al. 1987) and the Swiss Water balance Simulation Model ETH (WaSiM-ETH) (Schulla and Jasper 1999).

The best known model is the SHE, later developed into MIKE SHE, which simulates the entire land phase of the hydrologic cycle and allows components to be used independently and to be customized to local needs (Molnar et al. 2002). Conceptual models are relatively less complex and identify the structure of the model based only on selected component processes. One major type of this model is the simplified distributed model, which uses distribution functions to describe the spatial variability of surface runoff. Examples of such models are the TOPMODEL (Bevern and Kirkby 1979), Xinanjiang (Zhao et al. 1980), and the Probability Distributed Model (PDM) (Moore and Clarke 1981). A second type of conceptual model is the Explicit Soil Moisture Accounting (ESMA) model, which uses a number of conceptual reservoirs to describe subsurface water storage and transformation into discharge. Examples of ESMA models include the Sacramento Soil Moisture Accounting (SAC-SMA) (Burnash et al. 1973; Burnash 1995) and the Australian Large Scale Catchment Model (LASCAM) (Sivapalan et al. 1996a; b; c).

Since physically based and conceptual models are designed to mathematically simulate the physical mechanisms that determine the hydrological cycle, the model calibration is complex and requires a comprehensive knowledge of the studied basin. Metric models, on the other hand, are data-based and rely mostly on statistical estimations of observed data to describe the runoff responses. These models are usually based on time-series data and are inadequate for interpreting physical processes. Examples include the Box-Cox type discrete-time transfer functions (Box and Cox 1964) and neural network models (Bowden et al. 2005a; b), which have been applied successfully to forecast rainfall-runoff processes at different temporal stages, (e.g., Castellano-Méndez et al. 2004). The final category of models, the HMC, integrates the advantages of the statistical characterization of metric models and the prescribed physical interpretations of conceptual models, as transfer function models. They are able to represent the rainfall-runoff non-linearity and adequately characterize the physical processes of the flow mechanisms. Statistical procedures to assess the validity and quantify the uncertainties in parameter estimates and model outputs are also well developed (Young and Bevern 1994). Two main HMC approaches are the Deductive Approach and the Data-Based Mechanistic (DBM) model. The conceptual model is specified a priori in the deductive approach; the Identification of unit Hydrographs And Component flows from Rainfall, Evaporation and Streamflow data (IHACRES) model, for example, estimates only 6 to 7 parameters and has been widely applied in several regions and climatic conditions (e.g., Hansen et al. 1996; Littlewood and Marsh 1996). The second approach, DBM modeling, allows the data to suggest an appropriate model structure that is compatible with the available input-output information, and then evaluates the resulting model for mechanistic interpretation, (e.g., Mwakalila et al. 2001).

With the wide range of simple to complex process-driven hydrologic models available, researchers have compared the efficiency of numbers of competing models in order to select the most appropriate one (e.g., Lee et al. 2005; Clarke 2008; Das et al.

2008). The assessment of whether input data with high spatial resolution and high model resolution leads to improved model results for stream flow simulation, for example, has shown that more finely resolved input data and model resolution does not necessarily improve the model performance unless the input data corresponds to an increase of

information (Perrin et al. 2001; Das et al. 2008). How a model performs depends on several factors such as the scale of the catchment, physiographic characteristics of the catchments, availability of the data required to set up the selected model, the dominating rainfall type, seasonality of precipitation, season of the year and dominating runoff producing mechanisms. There is no single model that can perform consistently over the range of catchment types and conditions, and the choice of a model depends on the researcher’s preference and familiarity with a particular model, the objectives and available data. In some cases, instead of selecting a single model, the combination of results from several hydrologic models by the use of Bayesian statistical techniques can be advantageous in providing more accurate uncertainty estimates; although this can limit the usefulness of different models at different times (Niggli and Marsh 2005; Marshall et al. 2007). There are criticisms, however, that procedures used for comparison are not satisfactory because (i) they provide no measure of the uncertainty in model differences or model performance, (ii) validations and calibrations are usually carried out by persons familiar with the use of the model which results in less effectiveness for the general hydrological community, and (iii) there are relatively limited practices of good experimental design (i.e., replication and randomization) as compared to other fields of applied science (Clarke 2008).

Upland Streams

The catchment of a river system is conceptually divided into the headwater basins, the low-order stream system and the main river corridor, with the headwater basin as the area of most dynamic response to intense rainfall (White and Garci-Ruiz 1998). The topography, vegetative cover and soil properties are crucial variables for runoff production and erosion (Molnar et al. 2002). First- and second-order streams, whose total lengths form more than 70% of the total stream length in a river network, are important as they connect upland and riparian systems with river systems (Leopold et al. 1964). Their contribution to downstream stream flow (Winter 2007), nutrient cycling and watershed water quality (Alexander et al.

2007), biodiversity (Meyer et al. 2007), and organic material (Wipfli et al. 2007) is essential in watershed ecosystem functioning. The understanding of how lower-order streams influence water quality and flow of downstream systems is fundamental to the

effective management of all water resources. For example, first-order streams have been found to contribute to 70% of the mean annual water volume and 65% of nitrogen fluxes in second-order streams, and 55% and 40%, respectively, in fourth- and higher order rivers (Alexander et al. 2007). Since every important aspect of the river ecosystem, i.e., the river geomorphic system and the river chemical system, begins in lower-order streams, any change has the potential to reduce ecological integrity on much larger spatial scales (Freeman et al. 2007) .

Nutrient loads

The riverine ecosystem, as a dynamic system, connects the headwater streams to the landscape and transfers energy downstream to sites such as estuaries, coastal and marine ecosystem. The temporal and spatial variability in channel processes and features control how nutrient transfers along this longitudinal linkage undergo multiple cycles of uptake, biological, chemical and physical storage and re-mineralization – processes which are described by the nutrient spiraling concept (Newbold et al. 1981). Ecological models that have described processes in the natural river ecosystem include the River Continuum Concept (RCC) (Vannote et al. 1980), an earlier model that considers the river system as one ecological unit with physical and biological changes along the length of the river as a result of its terrestrial surroundings. The headstream, shaded by riparian forest canopies with poor light penetration, obtains energy sources mainly from riparian vegetation. The wider middle reaches, which are less shaded by vegetation, have increased primary productivity due to higher irradiation. The lower reaches, characterized by higher turbidity, depth and substratum instability, and reduced photosynthesis and energy inputs, mostly process organic material from upstream. The RCC, however, assumed a perennial river system based on temperate river systems with managed areas, and excluded the irregularities in small stream flows strongly influenced by local rainfall that resulted in breaks in the continuum. The Flood Pulse Concept (FPC) (Junk et al. 1989), although more applicable to larger rivers and the floodplains of downstream sites, suggested that periodic changes in the water level that led to lateral exchanges between river channel and the floodplain (the area on the sides of the channel that is inundated by floodwaters at intervals)

rather contributed to the energy sources of the river ecosystem. The FPC surmised that, in addition to the dissolved and suspended material from upstream, the nutrient status and associated aquatic fauna were influenced strongly by internal processes and nutrient transfer mechanisms caused by interaction with the terrestrial phase.

Increased contributions from land-use activities are either through a single source (point source), and thus generally easier to identify, quantify and control, or those that are from diffuse sources (non-point sources) which are more difficult to control. Fertilizer production, fossil fuel consumption and planting of leguminous crops, for example, have doubled the rate at which biologically available nitrogen enters the terrestrial biosphere (Galloway and Cowling 2004). Phosphate mining and its use in fertilizers have doubled phosphorus inputs over the natural weathering process (Bennett et al. 2001). There are multiple loss pathways for nutrients from fields to streams, and although there is much interest in understanding the nutrient loadings in soil in order to control their input into aquatic systems, the diffusion and delivery of these pollutants are independent of the source, and pathways, therefore, are not easily predictable.

One way of establishing nutrient loads in water bodies with respect to the land-use type, is by the use of an export coefficient, which estimates the mass of a specific nutrient exported from a particular land use in one year. Computer models have been developed to estimate pollutant loading from terrestrial sources into the water body, which range from simple export coefficient models (Arnold and Allen 1996), regression models such as SPARROW (Newham et al. 2004), to complex models such as SWAT (Matthies et al.

2006), CatchMODS (Dise 2004), and MONERIS (Sharpley et al. 2002). Export coefficient models estimate organic matter, nutrient sources and nutrient transport as a function of land area, with assumptions on export characteristics for a specific climatic regime and land-use types. The Integrated Nitrogen in Catchments (INCA) in Europe, for example, takes all nitrogen sources, processes them through crops, semi-natural vegetation, microbes and soil, to predict how much nitrogen will come out into the river or stream. Above 50 mg L^-1 which is the EU maximum, different options are used, such as cutting back on fertilizer application, changing crops or crop rotations, planting forests, shutting down or upgrading sewage works, or a combination of these strategies (McFarland and Hauck 2001). These

models are better at predicting annual averages than daily events or transient peaks that may come from unknown sources or poorly understood internal processes.

For smaller scale studies, field-scale nutrient load data are required to understand the nutrient transport mechanisms and variability in soil, land use, climate, topography and management (Wickam et al. 2003). Small watersheds with a predominant land use and field plots established to collect runoff from storm events are used for these investigations, although difficulties exist in identifying and monitoring catchments having a single homogenous land-use type (Sahoo et al. 2006), or data obtained might not represent the average conditions and practices such as soils, planting and harvesting dates, slopes, tillage practices, or proximity to streams (Duan et al. 2003). Partially understood physical processes that interact at complex non-linear scales and influence solute turnover in the soil, groundwater and stream water can also lead to underestimates of loads (Rousseau et al.

2002).

In some cases, more accurate estimates of nutrient losses from drainage basins are needed to predict responses of ecological systems to nutrient inputs; for instance, when direct estimation of pollutant loading is needed to either calibrate hydrological and water quality models (Letcher et al. 2002; Quilbé et al. 2006; Shrestha et al. 2008), or there is an interest in the load-based targets, i.e., the receiving water, where there are requirements to improve water quality such as the Total Maximum Daily Load (TMDL). Loads are estimated from measurements of pollutant concentrations in-stream and stream flow data, although adequate data are not always available due to the time, expense and effort required to collect and analyze field data. Generally, as continuous flow data are more readily obtainable than concentrations, several estimation techniques are available with guidelines to assess the most appropriate method based on the type of available data. Reviews carried out on various estimation techniques have shown that there are no universal methods available with precise or minimum variance, and little has been done about quantifying the uncertainties which surround the estimate loads (Littlewood 1992; Littlewood et al. 1998;

Mukhopadhyay and Smith 2000; Letcher et al. 2002; Etchells et al. 2005; Quilbé et al.

2006). Different methods can even give different results when applied to the same data set (Walling and Webb 1988).

Overall, there are three main classes of methods, with variations available in each depending on the type of available data: (i) averaging estimators or interpolation methods, (ii) ratio estimators, and (iii) regression methods. Averaging or interpolation techniques use the product of the sum of discharge volume and average concentration data sampled at different frequencies. They are generally biased when the period between sampling times increase, or if the data set does not represent the entire range of flows and concentration values; in such cases they are used for a first approximation (Dolan et al. 1981). The ratio estimator uses the available sampled concentration and flow time series to compute a fraction that accounts for the covariance between the two parameters. These estimators are considered unbiased when the relation between load and discharge is linear with an intercept of zero and when the variance of load is proportional to discharge. Both of these conditions are often approximately satisfied by the relationships between discharge and load, and the estimators are considered comparatively more statistically robust (Mukhopadhyay and Smith 2000; Vieux and Moreda 2003; Quilbé et al. 2006). Regression analyses generally perform well, but require a strong correlation between stream flow and nutrient concentration for a wide range of stream flow values, such that the concentration of non-sampled periods can be inferred from the flow data (Endreny and Hassett 2005; Johnes 2007). Three modifications are known, which include the quasi-maximum likelihood (QMLE) (Ferguson 1986), the non-parametric smearing estimator (Duan 1983), and the minimum variance unbiased estimator (MVUE) or Bradu-Mundlak estimator (Cohn et al.

1989; Cohn 1995). Fairly accurate estimates are obtained from the three modifications when (i) the assumed linear model is correct, (ii) the model is based on 30 or more observations, and (iii) the model does not extrapolate beyond the range of calibration data

In document The Influence of Land-use Activities on Nutrient Inputs into Upland Catchment Streams, Ghana (Page 38-200)